From: Panu Matilainen <pmatilai@redhat.com>
To: Keith Wiles <keith.wiles@intel.com>, dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v3] mempool: reduce rte_mempool structure size
Date: Fri, 12 Feb 2016 13:23:45 +0200 [thread overview]
Message-ID: <56BDC0C1.9010506@redhat.com> (raw)
In-Reply-To: <1455139134-9826-1-git-send-email-keith.wiles@intel.com>
On 02/10/2016 11:18 PM, Keith Wiles wrote:
> The rte_mempool structure is changed, which will cause an ABI change
> for this structure. Providing backward compat is not reasonable
> here as this structure is used in multiple defines/inlines.
>
> Allow mempool cache support to be dynamic depending on if the
> mempool being created needs cache support. Saves about 1.5M of
> memory used by the rte_mempool structure.
>
> Allocating small mempools which do not require cache can consume
> larges amounts of memory if you have a number of these mempools.
>
> Signed-off-by: Keith Wiles <keith.wiles@intel.com>
> ---
> * Patch v3 fix up the ifdefs to correct some problems in removing ifdef
> lines. Added the ABI deprecation notice to the document file.
> * Patch v2 to add some comments and setup for RTE_NEXT_ABI changes.
>
> app/test/test_mempool.c | 5 +++
> doc/guides/rel_notes/deprecation.rst | 7 +++
> lib/librte_mempool/rte_mempool.c | 82 +++++++++++++++++++++++++++++++++---
> lib/librte_mempool/rte_mempool.h | 46 ++++++++++++++++----
> 4 files changed, 127 insertions(+), 13 deletions(-)
>
> diff --git a/app/test/test_mempool.c b/app/test/test_mempool.c
> index f0f823b..f3fba50 100644
> --- a/app/test/test_mempool.c
> +++ b/app/test/test_mempool.c
> @@ -122,8 +122,13 @@ test_mempool_basic(void)
> return -1;
>
> printf("get private data\n");
> +#ifdef RTE_NEXT_ABI
> + if (rte_mempool_get_priv(mp) != (char *)mp +
> + MEMPOOL_HEADER_SIZE(mp, mp->pg_num, mp->cache_size))
> +#else
> if (rte_mempool_get_priv(mp) !=
> (char*) mp + MEMPOOL_HEADER_SIZE(mp, mp->pg_num))
> +#endif
> return -1;
>
> printf("get physical address of an object\n");
> diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
> index e94d4a2..1b9d25e 100644
> --- a/doc/guides/rel_notes/deprecation.rst
> +++ b/doc/guides/rel_notes/deprecation.rst
> @@ -49,3 +49,10 @@ Deprecation Notices
> commands (such as RETA update in testpmd). This should impact
> CMDLINE_PARSE_RESULT_BUFSIZE, STR_TOKEN_SIZE and RDLINE_BUF_SIZE.
> It should be integrated in release 2.3.
> +
> +* ABI change is planned for the rte_mempool structure to allow mempool
> + cache support to be dynamic depending on the mempool being created
> + needing cache support. Saves about 1.5M of memory per rte_mempool structure
> + by removing the per lcore cache memory. Change will occur after DPDK 16.04
> + release.
> +
> diff --git a/lib/librte_mempool/rte_mempool.c b/lib/librte_mempool/rte_mempool.c
> index aff5f6d..5f21eaa 100644
> --- a/lib/librte_mempool/rte_mempool.c
> +++ b/lib/librte_mempool/rte_mempool.c
> @@ -452,12 +452,17 @@ rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
> /* compilation-time checks */
> RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
> RTE_CACHE_LINE_MASK) != 0);
> +#ifdef RTE_NEXT_ABI
> + RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
> + RTE_CACHE_LINE_MASK) != 0);
> +#else
> #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_cache) &
> RTE_CACHE_LINE_MASK) != 0);
> RTE_BUILD_BUG_ON((offsetof(struct rte_mempool, local_cache) &
> RTE_CACHE_LINE_MASK) != 0);
> #endif
> +#endif /* RTE_NEXT_ABI */
> #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
> RTE_BUILD_BUG_ON((sizeof(struct rte_mempool_debug_stats) &
> RTE_CACHE_LINE_MASK) != 0);
> @@ -527,9 +532,8 @@ rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
> */
> int head = sizeof(struct rte_mempool);
> int new_size = (private_data_size + head) % page_size;
> - if (new_size) {
> + if (new_size)
> private_data_size += page_size - new_size;
> - }
> }
>
> /* try to allocate tailq entry */
> @@ -544,7 +548,12 @@ rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
> * store mempool objects. Otherwise reserve a memzone that is large
> * enough to hold mempool header and metadata plus mempool objects.
> */
> +#ifdef RTE_NEXT_ABI
> + mempool_size = MEMPOOL_HEADER_SIZE(mp, pg_num, cache_size);
> + mempool_size += private_data_size;
> +#else
> mempool_size = MEMPOOL_HEADER_SIZE(mp, pg_num) + private_data_size;
> +#endif /* RTE_NEXT_ABI */
> mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
> if (vaddr == NULL)
> mempool_size += (size_t)objsz.total_size * n;
> @@ -598,9 +607,22 @@ rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
> mp->cache_flushthresh = CALC_CACHE_FLUSHTHRESH(cache_size);
> mp->private_data_size = private_data_size;
>
> +#ifdef RTE_NEXT_ABI
> + /*
> + * local_cache pointer is set even if cache_size is zero.
> + * The local_cache points to just past the elt_pa[] array.
> + */
> + mp->local_cache = (struct rte_mempool_cache *)
> + ((char *)mp + MEMPOOL_HEADER_SIZE(mp, pg_num, 0));
> +
> + /* calculate address of the first element for continuous mempool. */
> + obj = (char *)mp + MEMPOOL_HEADER_SIZE(mp, pg_num, cache_size) +
> + private_data_size;
> +#else
> /* calculate address of the first element for continuous mempool. */
> obj = (char *)mp + MEMPOOL_HEADER_SIZE(mp, pg_num) +
> private_data_size;
> +#endif /* RTE_NEXT_ABI */
> obj = RTE_PTR_ALIGN_CEIL(obj, RTE_MEMPOOL_ALIGN);
>
> /* populate address translation fields. */
> @@ -613,9 +635,8 @@ rte_mempool_xmem_create(const char *name, unsigned n, unsigned elt_size,
> mp->elt_va_start = (uintptr_t)obj;
> mp->elt_pa[0] = mp->phys_addr +
> (mp->elt_va_start - (uintptr_t)mp);
> -
> - /* mempool elements in a separate chunk of memory. */
> } else {
> + /* mempool elements in a separate chunk of memory. */
> mp->elt_va_start = (uintptr_t)vaddr;
> memcpy(mp->elt_pa, paddr, sizeof (mp->elt_pa[0]) * pg_num);
> }
> @@ -645,10 +666,21 @@ unsigned
> rte_mempool_count(const struct rte_mempool *mp)
> {
> unsigned count;
> +#ifdef RTE_NEXT_ABI
> + unsigned lcore_id;
>
> count = rte_ring_count(mp->ring);
>
> + if (mp->cache_size == 0)
> + return count;
> +
> + for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
> + count += mp->local_cache[lcore_id].len;
> +#else
> #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> +
> + count = rte_ring_count(mp->ring);
> +
> {
> unsigned lcore_id;
> if (mp->cache_size == 0)
> @@ -658,6 +690,7 @@ rte_mempool_count(const struct rte_mempool *mp)
> count += mp->local_cache[lcore_id].len;
> }
> #endif
> +#endif /* RTE_NEXT_ABI */
>
> /*
> * due to race condition (access to len is not locked), the
> @@ -672,6 +705,24 @@ rte_mempool_count(const struct rte_mempool *mp)
> static unsigned
> rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
> {
> +#ifdef RTE_NEXT_ABI
> + unsigned lcore_id;
> + unsigned count = 0;
> + unsigned cache_count;
> +
> + fprintf(f, " cache infos:\n");
> + fprintf(f, " cache_size=%"PRIu32"\n", mp->cache_size);
> + if (mp->cache_size == 0)
> + return count;
> +
> + for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
> + cache_count = mp->local_cache[lcore_id].len;
> + fprintf(f, " cache_count[%u]=%u\n", lcore_id, cache_count);
> + count += cache_count;
> + }
> + fprintf(f, " total_cache_count=%u\n", count);
> + return count;
> +#else
> #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> unsigned lcore_id;
> unsigned count = 0;
> @@ -691,6 +742,7 @@ rte_mempool_dump_cache(FILE *f, const struct rte_mempool *mp)
> fprintf(f, " cache disabled\n");
> return 0;
> #endif
> +#endif /* RTE_NEXT_ABI */
> }
>
> #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
> @@ -755,6 +807,26 @@ mempool_audit_cookies(const struct rte_mempool *mp)
> #define mempool_audit_cookies(mp) do {} while(0)
> #endif
>
> +#ifdef RTE_NEXT_ABI
> +/* check cookies before and after objects */
> +static void
> +mempool_audit_cache(const struct rte_mempool *mp)
> +{
> + /* check cache size consistency */
> + unsigned lcore_id;
> +
> + if (mp->cache_size == 0)
> + return;
> +
> + for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
> + if (mp->local_cache[lcore_id].len > mp->cache_flushthresh) {
> + RTE_LOG(CRIT, MEMPOOL, "badness on cache[%u]\n",
> + lcore_id);
> + rte_panic("MEMPOOL: invalid cache len\n");
> + }
> + }
> +}
> +#else
> #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> /* check cookies before and after objects */
> static void
> @@ -773,7 +845,7 @@ mempool_audit_cache(const struct rte_mempool *mp)
> #else
> #define mempool_audit_cache(mp) do {} while(0)
> #endif
> -
> +#endif /* RTE_NEXT_ABI */
>
> /* check the consistency of mempool (size, cookies, ...) */
> void
> diff --git a/lib/librte_mempool/rte_mempool.h b/lib/librte_mempool/rte_mempool.h
> index 9745bf0..b12d6a9 100644
> --- a/lib/librte_mempool/rte_mempool.h
> +++ b/lib/librte_mempool/rte_mempool.h
> @@ -95,7 +95,7 @@ struct rte_mempool_debug_stats {
> } __rte_cache_aligned;
> #endif
>
> -#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> +#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0 || defined(RTE_NEXT_ABI) /* Remove line */
> /**
> * A structure that stores a per-core object cache.
> */
> @@ -107,7 +107,7 @@ struct rte_mempool_cache {
> */
> void *objs[RTE_MEMPOOL_CACHE_MAX_SIZE * 3]; /**< Cache objects */
> } __rte_cache_aligned;
> -#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */
> +#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */ /* Remove line RTE_NEXT_ABI */
>
> /**
> * A structure that stores the size of mempool elements.
> @@ -194,10 +194,14 @@ struct rte_mempool {
>
> unsigned private_data_size; /**< Size of private data. */
>
> +#ifdef RTE_NEXT_ABI
> + struct rte_mempool_cache *local_cache; /**< Per-lcore local cache */
> +#else
> #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> /** Per-lcore local cache. */
> struct rte_mempool_cache local_cache[RTE_MAX_LCORE];
> #endif
> +#endif /* RTE_NEXT_ABI */
>
> #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
> /** Per-lcore statistics. */
> @@ -246,6 +250,26 @@ struct rte_mempool {
> #define __MEMPOOL_STAT_ADD(mp, name, n) do {} while(0)
> #endif
>
> +#ifdef RTE_NEXT_ABI
> +/**
> + * Size of elt_pa array size based on number of pages. (Internal use)
> + */
> +#define __PA_SIZE(mp, pgn) \
> + RTE_ALIGN_CEIL((((pgn) - RTE_DIM((mp)->elt_pa)) * \
> + sizeof((mp)->elt_pa[0])), RTE_CACHE_LINE_SIZE)
> +
> +/**
> + * Calculate the size of the mempool header.
> + *
> + * @param mp
> + * Pointer to the memory pool.
> + * @param pgn
> + * Number of pages used to store mempool objects.
> + */
> +#define MEMPOOL_HEADER_SIZE(mp, pgn, cs) \
> + (sizeof(*(mp)) + __PA_SIZE(mp, pgn) + (((cs) == 0) ? 0 : \
> + (sizeof(struct rte_mempool_cache) * RTE_MAX_LCORE)))
> +#else
> /**
> * Calculate the size of the mempool header.
> *
> @@ -257,6 +281,7 @@ struct rte_mempool {
> #define MEMPOOL_HEADER_SIZE(mp, pgn) (sizeof(*(mp)) + \
> RTE_ALIGN_CEIL(((pgn) - RTE_DIM((mp)->elt_pa)) * \
> sizeof ((mp)->elt_pa[0]), RTE_CACHE_LINE_SIZE))
> +#endif /* RTE_NEXT_ABI */
>
> /**
> * Return true if the whole mempool is in contiguous memory.
> @@ -755,19 +780,19 @@ static inline void __attribute__((always_inline))
> __mempool_put_bulk(struct rte_mempool *mp, void * const *obj_table,
> unsigned n, int is_mp)
> {
> -#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> +#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0 || defined(RTE_NEXT_ABI) /* Remove line */
> struct rte_mempool_cache *cache;
> uint32_t index;
> void **cache_objs;
> unsigned lcore_id = rte_lcore_id();
> uint32_t cache_size = mp->cache_size;
> uint32_t flushthresh = mp->cache_flushthresh;
> -#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */
> +#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */ /* Remove line RTE_NEXT_ABI */
>
> /* increment stat now, adding in mempool always success */
> __MEMPOOL_STAT_ADD(mp, put, n);
>
> -#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> +#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0 || defined(RTE_NEXT_ABI) /* Remove line */
> /* cache is not enabled or single producer or non-EAL thread */
> if (unlikely(cache_size == 0 || is_mp == 0 ||
> lcore_id >= RTE_MAX_LCORE))
> @@ -802,7 +827,7 @@ __mempool_put_bulk(struct rte_mempool *mp, void * const *obj_table,
> return;
>
> ring_enqueue:
> -#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */
> +#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */ /* Remove line RTE_NEXT_ABI */
>
> /* push remaining objects in ring */
> #ifdef RTE_LIBRTE_MEMPOOL_DEBUG
> @@ -946,7 +971,7 @@ __mempool_get_bulk(struct rte_mempool *mp, void **obj_table,
> unsigned n, int is_mc)
> {
> int ret;
> -#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0
> +#if RTE_MEMPOOL_CACHE_MAX_SIZE > 0 || defined(RTE_NEXT_ABI) /* Remove line */
> struct rte_mempool_cache *cache;
> uint32_t index, len;
> void **cache_objs;
> @@ -992,7 +1017,7 @@ __mempool_get_bulk(struct rte_mempool *mp, void **obj_table,
> return 0;
>
> ring_dequeue:
> -#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */
> +#endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */ /* Remove line RTE_NEXT_ABI */
>
> /* get remaining objects from ring */
> if (is_mc)
> @@ -1293,7 +1318,12 @@ void rte_mempool_audit(const struct rte_mempool *mp);
> */
> static inline void *rte_mempool_get_priv(struct rte_mempool *mp)
> {
> +#ifdef RTE_NEXT_ABI
> + return (char *)mp +
> + MEMPOOL_HEADER_SIZE(mp, mp->pg_num, mp->cache_size);
> +#else
> return (char *)mp + MEMPOOL_HEADER_SIZE(mp, mp->pg_num);
> +#endif /* RTE_NEXT_ABI */
> }
>
> /**
>
This is not RTE_NEXT_ABI material IMO, the added ifdef clutter is just
too much.
I'd suggest adding a deprecation notice for the change now and after
16.04 is released, just resend the patch without messing with RTE_NEXT_ABI.
- Pnau -
next prev parent reply other threads:[~2016-02-12 11:23 UTC|newest]
Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-02-02 23:02 [dpdk-dev] [PATCH] mempool: Reduce " Keith Wiles
2016-02-03 17:11 ` Ananyev, Konstantin
2016-02-08 11:02 ` Olivier MATZ
2016-02-08 15:57 ` Wiles, Keith
2016-02-09 17:30 ` [dpdk-dev] [PATCH v2] mempool: reduce " Keith Wiles
2016-02-10 16:59 ` Olivier MATZ
2016-02-10 17:22 ` Wiles, Keith
2016-02-10 18:35 ` Wiles, Keith
2016-02-10 20:06 ` Olivier MATZ
2016-02-10 21:18 ` [dpdk-dev] [PATCH v3] " Keith Wiles
2016-02-12 11:23 ` Panu Matilainen [this message]
2016-02-12 13:57 ` Thomas Monjalon
2016-02-12 14:19 ` Panu Matilainen
2016-02-12 15:07 ` Wiles, Keith
2016-02-12 15:38 ` Thomas Monjalon
2016-02-12 15:50 ` Olivier MATZ
2016-02-12 15:58 ` Wiles, Keith
2016-02-15 9:58 ` Hunt, David
2016-02-15 10:15 ` Olivier MATZ
2016-02-15 10:21 ` Hunt, David
2016-02-15 12:31 ` Olivier MATZ
2016-02-12 15:54 ` Wiles, Keith
2016-02-12 18:36 ` [dpdk-dev] [PATCH v4] " Keith Wiles
2016-02-15 9:20 ` Olivier MATZ
2016-04-14 9:42 ` [dpdk-dev] [PATCH v5] " Olivier Matz
2016-04-14 13:28 ` Wiles, Keith
2016-04-14 13:43 ` Olivier MATZ
2016-04-14 13:53 ` Wiles, Keith
2016-05-17 5:31 ` Thomas Monjalon
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=56BDC0C1.9010506@redhat.com \
--to=pmatilai@redhat.com \
--cc=dev@dpdk.org \
--cc=keith.wiles@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).